10Web test data

The café brief was typed word for word into 10Web's prompt box on its dashboard. The clock ran from pressing Enter to the moment the five pages opened at the public address.

10Web did not ask questions. About 7 seconds after Enter it showed a written plan (a goal, a design direction with colours, the pages) and the button 'Approve & start generation'. I pressed it. The five pages, a logo and 26 pictures were live about 7 minutes after the brief, and the credit meter still read 0 of 5.

It built the whole site in one go, so no second message was needed. The name, the type of business, the address and the hours were on the draft.

• Times are from my own session notes. About 2 to 3 of the 7 minutes were the plan waiting for my click.

Each of the twelve points was tried as far as the trial allowed.

Domain: the Publish screen offers three names marked 'Free' (juniper-rye.com, juniperryefarm.com, juniperrye.co.uk) under 'Free-up to $30', and a button 'Connect your own domain'. I did not buy or connect one. The pricing page says the free domain comes with a paid plan.

Form: the café contact form has name, email, phone, a topic, an event date and a message, with a hidden spam trap field. A required 'Party size' field added by the AI showed on the live form. Entries are listed under Forms in the admin. I did not send a test message.

The WordPress Plugins screen answered 'Sorry, you are not allowed to access this page.' A shop, a blog post, a mailbox, a test order and tax were not tried.

• No card was entered.

I opened the WordPress export screen and downloaded an export of the whole site, and read the plan screen for what happens when the trial ends.

The WordPress Export screen produced a 5.5 MB XML file in one click, with 7 page items and 26 pictures. The in-app plan screen says 'No charge after 3 days unless you choose to upgrade'. The pricing page's own FAQ says the site is deleted when the trial ends unless a plan is picked. I could not test cancelling a paid plan.

• I did not import the file into another WordPress site.

I emailed three questions to support@10web.io on 28 October 2026 and typed five searches into the help centre.

No reply had arrived when I wrote this. The help centre's search page asked me to pass a human check, which I did not, so I used its public search feed. Two of five searches had a correct answer in the top three results: adding custom code (an article on header codes) and changing the billing card (the Subscription Plan page, which lists Payment methods). Connecting a domain bought elsewhere, form messages not arriving and exporting a site found no answering article in the top three.

• Live chat was not used.

All five pages were loaded at 360, 768 and 1440 pixels wide on the trial address. The checker counted sideways scrolling, tap targets under 24 pixels, text under WCAG AA contrast, placeholders, broken images and overlapping text, once per type per page. I also ran axe-core on every page at 1440 and 390 pixels and walked the home page with the Tab key.

7 faults on 5 pages: tap targets under 24 pixels on all 5 pages (small footer and menu links) and text that is too pale on 2 pages. No sideways scrolling, no placeholders, no broken images and no overlapping text.

axe-core found no critical problem. It found one serious problem on every page (a link with no readable name) and one more on the menu page (low contrast), plus moderate ones about page landmarks. The page language is set and no image lacks alt text (35 of 35). With the Tab key I reached 8 to 20 controls per page and every focus ring showed except on one control on the home page.

Template switching (WB-D1): the theme panel offers 10 ready-made themes and one click on 'Fresh lime' restyled the preview at once. I discarded it before saving, so I did not count how many facts survived.

• One site only, so the look-alike check across three drafts was not run.

The 20 café facts were checked by hand on the published pages, and every claim the brief did not give was counted.

All 20 facts were right: the name, owner, 2019, address, phone, email, both sets of hours, all five prices, oat milk, Wi-Fi, the patio, catering for 40 people with 48 hours' notice, the Instagram name and the five pages with a contact form.

10Web also added 12 small claims I never gave: wild honey and French lavender in the latte, 'mixed, shaped and pulled by hand', quiet tables with power sockets, a cedar patio with shade umbrellas, a 36-hour rye loaf, a Swedish-style cardamom bun, an egg that is 'pasture-raised', 'rotating single-origin' coffee, 'gluten-conscious options', stone hearths, 'fresh pastries' for catering and an origin story. It added no reviews, awards, staff or other social networks.

• Only one of the three briefs was run, so this is 20 facts, not 60.

Twelve fixed edits were made on the live café site. Seven were made by hand in the visual editor and five by asking the AI chat. Each was checked on the public address after publishing.

10 of 12 worked. Moving a section, new colours, a new Gift Cards page with a menu link, a footer line, a required form field, a replaced picture, a button link, a duplicated section, an embedded HTML snippet and a warmer About paragraph all appeared on the public address.

Two did not work as asked. The site font changed the body text but not the headings. A heading size set while the editor showed the phone preview applied at every width, so the phone-only change leaked to the desktop. Dragging a section in the panel did not drop, so the AI moved it.

Each hand edit took 3 to 6 clicks. Each AI edit took about one minute and cost 1 of the 5 trial credits. Every change needs Save and then Publish changes, about 10 to 15 seconds.

• The blank-page build, the gallery, slider and video embed, and the phone app were not run.

Lighthouse 13.5 with mobile settings ran three times on each of the five pages (15 runs), and three times on the home page with the desktop setting, on the trial address.

The median of the 15 mobile runs was 74. By page: home 66, menu 79, about 76, catering 71, contact 78. The home page was slowest: its main picture showed after 7.0 seconds on the test phone because the page weighs about 3.1 MB. The desktop median on the home page was 84.

Layout shift was 0 and blocking time was 0 to 30 ms. Most photos are PNG files of 480 to 730 KB; only 4 of 35 images use a modern format, and 30 are larger than they are shown. The server answered in about 0.4 seconds.

• Measured on the trial address from one place; speed from three regions was not run.

Eleven of the twelve points were read in the source code and files of the live pages. Heading order was not checked.

Present on every page: a title, a meta description, one H1 in the page source, a canonical tag, Open Graph and Twitter card tags, and structured data (WebSite, Organization, WebPage, BreadcrumbList). Every image has alt text with a readable file name, and the page addresses are clean. A sitemap index with two sitemaps and a robots.txt exist, and a Redirects screen and a search-engine switch are in the WordPress admin.

axe-core said the menu and catering pages have no visible top-level heading. The site was open to search engines when it was built; I switched on WordPress's 'Discourage search engines' setting, and the pages then sent noindex. robots.txt names none of the AI crawlers I check.

• Google Search Console and Bing were not run.

Every plan was read on the pricing page and on the in-app upgrade screen, and the trial's limits on the plan screen.

The trial: 1 website, 1 website generation, 5 AI credits (any AI request costs 1), 5 GB of storage, 9,000 visitors, 3 days. AI Starter is $20 a month billed monthly or $10 a month billed yearly ('Get 12 months for $120 (regular price $240)'), with 1 website, 100 AI credits a month, 10 GB, 10,000 visitors and a free domain worth up to $30 for the first year. AI Premium is $15 a month yearly (2 websites, 200 credits). AI Ultimate is $22.50 a month yearly (4 websites, 300 credits). The pricing page promises a 30-day money-back guarantee.

The build used 0 credits and each edit used 1, so Starter's 100 credits cover about 100 edits a month, which is a few full sites like this one. The page does not say what the yearly plan costs at renewal; the regular price it shows is $240 a year.

• I did not buy a plan, so the renewal price and the billing page were not seen.